feat: Add LiteLLM as a direct connection provider

[RheagalFire]

Hi @OvidijusParsiunas,

I'd like to propose adding LiteLLM as a new direct connection provider. LiteLLM is an AI gateway that provides a unified OpenAI-compatible API for 100+ LLM providers (OpenAI, Anthropic, Azure, AWS Bedrock, Vertex AI, Groq, Ollama, etc.).

Why this would be useful:

• Users who self-host their LLM infrastructure can route through a single LiteLLM proxy instead of configuring individual providers
• Access to providers not yet directly supported by deep-chat (e.g., AWS Bedrock, Vertex AI, Azure OpenAI)
• Built-in provider routing, load balancing, and cost tracking

Proposed implementation:

• New liteLLM option in directConnection, following the same pattern as DeepSeek/OpenRouter
• Configurable proxy URL (default http://localhost:4000/v1/chat/completions)
• Standard OpenAI-compatible request/response format (which is what LiteLLM proxy serves)
• Key verification via the proxy's /v1/models endpoint

Example usage:

<deep-chat
  directConnection='{
    "liteLLM": {
      "key": "sk-your-litellm-key",
      "url": "http://localhost:4000/v1/chat/completions",
      "model": "anthropic/claude-3-5-sonnet"
    }
  }'
></deep-chat>

I already have an implementation ready following the existing provider pattern.

[OvidijusParsiunas]

Hi @RheagalFire.
Thankyou for your contribution! I have merged the code.

After looking at it in your detail I wanted to ask why is the API "key" in the interface. To my understanding this behaves similarly like Ollama where the core functionality is hosted on.

Developers will of-course host LiteLLM in their own servers with their own configurations, and for when that is the case we recommend them to use interceptors to fit their use-cases. Hence, for this reason I have removed the "key" interface, however if there is a unique peace in the API documentation that I should know about, please let me know. Thankyou!

This is currently available in deep-chat-dev and deep-chat-react-dev version 9.0.369.

I will close this issue when this will be published to the core packages!

[RheagalFire]

@OvidijusParsiunas thanks for merging!

To answer your question - yes, there is a difference from Ollama. LiteLLM proxy requires a auth key in production deployments. When set, all requests need an Authorization: Bearer <key> header or the proxy returns 401.

This is shown in the deploy docs: https://docs.litellm.ai/docs/proxy/deploy.

So while a local dev setup might run without auth (like Ollama), most production deployments will need the key.

[OvidijusParsiunas]

Thankyou for the feedback.
This all makes sense.

I have added the key property back into the API.
This is available in deep-chat-dev and deep-chat-react-dev version 9.0.370.

Thankyou!